US 20060156303 A1
A method of processing network data in a network processor includes using three or more threads to process a beginning portion, a middle portion, and an end portion of data packet. The first thread processes the beginning portion; one or more middle threads process the middle portion, and a last thread processes the end portion. First information is indirectly passed from the first thread to the last thread via a first buffer with the middle threads progressively updating the first information. Second information is directly passed from the first thread to the last thread via a second buffer.
1. A method of processing network data comprising:
using three or more threads of a set of threads provided by multiple programmable multi-threaded engines integrated within a processor to process a beginning portion, a middle portion, and an end portion of a data packet, wherein a first thread processes the beginning portion of the data packet, one or more middle threads process the middle portion of the data packet, and a last thread processes the end portion of the data packet;
passing first information from the first thread indirectly to the last thread via a first buffer, wherein each of the one or more middle threads progressively updates the first information;
passing second information from the first thread directly to the last thread via a second buffer.
2. The method of
3. The method of
4. The method of
5. The method of
6. The method of
7. The method of
8. The method of
9. A method of processing a data packet comprising:
processing a first portion of the data packet by a first thread of a set of threads provided by multiple programmable multi-threaded engines integrated within a processor;
processing a middle portion of the data packet by a second thread of the set of threads provided by the multiple programmable multi-threaded engines integrated within the processor; and
processing an end portion of the data packet by a final thread of the set of threads provided by the multiple programmable multi-threaded engines integrated within the processor;
storing first information by the first thread to be retrieved by the second thread;
storing second information by the first thread to be retrieved by the final thread.
10. The method of
11. The method of
12. The method of
13. The method of
placing the data packet in an output queue for sending out a port based on the second information.
14. The method of
storing second status information by the second thread;
processing a second middle portion of the data packet by a third thread; and
retrieving the second status information by the third thread.
15. A method of processing a data packet comprising:
using a first thread of a set of threads provided by multiple programmable multi-threaded engines integrated within a processor to process a first section of the data packet, the first thread saving state information, storing the first section of the data packet to a memory and saving output port information;
using two or more middle threads of the set of threads provided by the multiple programmable multi-threaded engines integrated within the processor to process a middle section of the data packet, each of the two or more middle threads retrieving the state information, saving updated state information, and storing at least a portion of the middle section of the data packet to the memory; and
using a final thread of the set of threads provided by the multiple programmable multi-threaded engines integrated within the processor to process a last section of the data packet, the final thread retrieving the state information, and storing the last section of the data packet to the memory.
17. The method of
using a final thread to process a last section of the data comprises retrieving the output port information.
18. The method of
queueing the data packet by the final thread to be transmitted based on the output port information.
19. The method of
20. An article comprising a computer-readable medium which store computer-executable instructions for receiving data from a plurality of ports, the instructions causing a computer to:
use a first thread of a set of threads provided by multiple programmable multi-threaded engines integrated within a processor to process a first section of a data packet including saving state information, storing the first section of the data packet to a memory, and saving output port information;
use a middle thread of the set of threads provided by the multiple programmable multi-threaded engines integrated within the processor to process a middle section of the data packet including retrieving the state information, incrementing an element counter, saving updated state information, and storing the middle section of the data packet to the memory; and
use a final thread of the set of threads provided by the multiple programmable multi-threaded engines integrated within the processor to process a last section of the data packet including retrieving the state information, incrementing the element counter, retrieving the output port information, and storing the last section of the data packet to the memory.
21. The article of
This application is a continuation of, and claims the priority entitled to U.S. patent application Ser. No. 09/710,439, filed Nov. 10, 2000, titled “MULTI-THREADED SEQUENCED RECEIVE FOR FAST NETWORK PORT STREAM OF PACKETS”, which is, in turn, a continuation-in-part of U.S. Pat. Ser. No. 09/626,535, filed Oct. 27, 2000, titled “MULTI-THREADED SCHEDULED RECEIVE FOR FAST NETWORK PORT DATA” which is, in turn, a continuation-in-part of U.S. Pat. Ser. No. 09/475,614, filed Dec. 30, 1999, titled “METHOD AND APPARATUS FOR CONTROL OF RECEIVE DATA”.
1. Field of the Invention
The described invention relates to the field of network communications. In particular, the invention relates to a method for using multiple threads to process incoming network data.
2. Description of Related Art
Networking products such as routers require high speed components for packet data movement, i.e., collecting packet data from incoming network device ports and queuing the packet data for transfer to appropriate forwarding device ports. They also require high-speed special controllers for processing the packet data, that is, parsing the data and making forwarding decisions. Because the implementation of these high-speed functions usually involves the development of ASIC or custom devices, such networking products are of limited flexibility. For example, each controller is assigned to service network packets from one or more given ports on a permanent basis.
A method of using multiple receive threads to receive data in a round robin scheme is disclosed. First, one embodiment of the hardware that enables the multiple thread system is described. Then the method of employing the multiple threads in a round robin fashion is described.
The hardware-based multi-threaded processor 12 also includes a processor 23 that assists in loading microcode control for other resources of the hardware-based multi-threaded processor 12 and performs other general purpose computer type functions such as handling protocols, exceptions, extra support for packet processing where the microengines pass the packets off for more detailed processing. In one embodiment, the processor 23 is a StrongARM (ARM is a trademark of ARM Limited, United Kingdom) core based architecture. The processor (or core) 23 has an operating system through which the processor 23 can call functions to operate on the microengines 22 a-22 f. The processor 23 can use any supported operating system, preferably real-time operating system. For the core processor implemented as a StrongARM architecture, operating systems such as MicrosoftNT real-time, VXWorks and :CUS, a freeware operating system available over the Internet, can be used.
The six microengines 22 a-22 f each operate with shared resources including the memory system 18, a PCI bus interface 24 and an I/O bus interface 28. The PCI bus interface provides an interface to the PCI bus 14. The I/O bus interface 28 is responsible for controlling and interfacing the processor 12 to the I/O bus 16. The memory system 18 includes a Synchronous Dynamic Random Access Memory (SDRAM) 18 a, which is accessed via an SDRAM controller 26 a, a Static Random Access Memory (SRAM) 18 b, which is accessed using an SRAM controller 26 b, and a nonvolatile memory (shown as a FlashROM) 18 c that is used for boot operations. The SDRAM 16 a and SDRAM controller 26 a are typically used for processing large volumes of data, e.g., processing of payloads from network packets. The SRAM 18 b and SRAM controller 26 b are used in a networking implementation for low latency, fast access tasks, e.g., accessing look-up tables, memory for the processor 23, and so forth. The microengines 22 a-22 f can execute memory reference instructions to either the SDRAM controller 26 a or the SRAM controller 18 b.
The hardware-based multi-threaded processor 12 interfaces to network devices such as a media access controller (“MAC”) device, including a “slow” device 30 (e.g., 10/100 BaseT Ethernet MAC) and/or a “fast” device 31, such as Gigabit Ethernet MAC, ATM device or the like, over the I/O Bus 16. In the embodiment shown, the slow device 30 is an 10/100 BaseT Octal MAC device and thus includes 8 slow ports 32 a-32 h, and the fast device is a Dual Gigabit MAC device having two fast ports 33 a, 33 b. Each of the network devices attached to the I/O Bus 16 can include a plurality of ports to be serviced by the processor 12. Other devices, such as a host computer (not shown), that may be coupled to the PCI bus 14 are also serviced by the processor 12. In general, as a network processor, the processor 12 can interface to any type of communication device or interface that receives/sends large amounts of data. The processor 12 functioning as a network processor could receive units of packet data from the devices 30, 31 and process those units of packet data in a parallel manner, as will be described. The unit of packet data could include an entire network packet (e.g., Ethernet packet) or a portion of such a packet.
Each of the functional units of the processor 12 are coupled to one or more internal buses. The internal buses include an internal core bus 34 (labeled “AMBA”) for coupling the processor 23 to the memory controllers 26 a, 26 b and to an AMBA translator 36. The processor 12 also includes a private bus 38 that couples the microengines 22 a-22 f to the SRAM controller 26 b, AMBA translator 36 and the Fbus interface 28. A memory bus 40 couples the memory controllers 26 a, 26 b to the bus interfaces 24, 28 and the memory system 18.
In addition to event signals that are local to an executing thread, the microengine employs signaling states that are global. With signaling states, an executing thread can broadcast a signal state to all microengines 22. Any and all threads in the microengines can branch on these signaling states. These signaling states can be used to determine availability of a resource or whether a resource is due for servicing.
The context event logic 74 has arbitration for the four threads. In one embodiment, the arbitration is a round robin mechanism. However, other arbitration techniques, such as priority queuing or weighted fair queuing, could be used. The microengine 22 a also includes an execution box (EBOX) data path 76 that includes an arithmetic logic unit (ALU) 76 a and a general purpose register (GPR) set 76 b. The ALU 76 a performs arithmetic and logical functions as well as shift functions.
The microengine 22 a further includes a write transfer register file 78 and a read transfer register file 80. The write transfer register file 78 stores data to be written to a resource. The read transfer register file 80 is for storing return data from a resource. Subsequent to or concurrent with the data arrival, an event signal from the respective shared resource, e.g., memory controllers 26 a, 26 b, or core 23, will be provided to the context event arbiter 74, which in turn alerts the thread that the data is available or has been sent. Both transfer register files 78, 80 are connected to the EBOX 76 through a data path. In the described implementation, each of the register files includes 64 registers.
The functionality of the microengine threads is determined by microcode loaded (via the core processor) for a particular user's application into each microengine's control store 70. Referring to
The receive scheduler thread 92 assigns packets to receive processing threads 96. In a packet forwarding application for a bridge/router, for example, the receive processing thread parses packet headers and performs lookups based in the packet header information. Once the receive processing thread or threads 96 has processed the packet, it either sends the packet as an exception to be further processed by the core 23 (e.g., the forwarding information cannot be located in lookup and the core processor must learn it), or stores the packet in the SDRAM and queues the packet in a transmit queue by placing a packet link descriptor for it in a transmit queue associated with the transmit (forwarding port) indicated by the header/lookup. The transmit queue is stored in the SRAM. The transmit arbiter thread 100 prioritizes the transmit queues and the transmit scheduler thread 94 assigns packets to transmit processing threads that send the packet out onto the forwarding port indicated by the header/lookup information during the receive processing.
The receive processing threads 96 may be dedicated to servicing particular ports or may be assigned to ports dynamically by the receive scheduler thread 92. For certain system configurations, a dedicated assignment may be desirable. For example, if the number of ports is equal to the number of receive processing threads 96, then it may be quite practical as well as efficient to assign the receive processing threads to ports in a one-to-one, dedicated assignment. In other system configurations, a dynamic assignment may provide a more efficient use of system resources.
The receive scheduler thread 92 maintains scheduling information 104 in the GPRs 76 b of the microengine within which it executes. The scheduling information 104 includes thread capabilities information 106, port-to-thread assignments (list) 108 and “thread busy” tracking information 110. At minimum, the thread capabilities information informs the receive scheduler thread as to the type of tasks for which the other threads are configured, e.g., which threads serve as receive processing threads. Additionally, it may inform the receive scheduler of other capabilities that may be appropriate to the servicing of a particular port. For instance, a receive processing thread may be configured to support a certain protocol, or a particular port or ports. A current list of the ports to which active receive processing threads have been assigned by the receive scheduler thread is maintained in the thread-to-port assignments list 108. The thread busy mask register 110 indicates which threads are actively servicing a port. The receive scheduler uses all of this scheduling information in selecting threads to be assigned to ports that require service for available packet data, as will be described in further detail below.
The processors 23 and 22 issue commands to the push/pull engine interface 122 when accessing one of the resources 120. The push/pull engine interface 122 places the commands into queues (not shown), arbitrates which commands to service, and moves data between the resources 120, the core 23 and the microengines 22. In addition to servicing requests from the core 23 and microengines 22, the push/pull engines 122 also service requests from the ready bus 128 to transfer control information to a register in the microengine read transfer registers 80.
When a thread issues a request to a resource 120, a command is driven onto an internal command bus 150 and placed in queues within the push/pull engine interface 122. Receive/read-related instructions (such as instructions for reading the CSRs) are written to a “push” command queue.
The CSRs 138 include the following types of registers: Fbus receive and transmit registers; Fbus and ready bus configuration registers; ready bus control registers; hash unit configuration registers; interrupt registers; and several miscellaneous registers, including a thread status registers. Those of the registers which pertain to the receive process will be described in further detail.
The interrupt/signal registers include an INTER_THD_SIG register for inter-thread signaling. Any thread within the microengines 22 or the core 23 can write a thread number to this register to signal an inter-thread event.
Further details of the Fbus controller 130 and the ready bus controller 126 are shown in
Some of the relevant CSRs used to program and control the ready bus 128 and Fbus 132 for receive processes are shown in
Although not depicted in detail, other bus registers include the following: a RDYBUS_TEMPLATE_CTL register 178 (
When the ready bus sequencer is programmed with an appropriate instruction directing it to interrogate MAC receive ready flags, it reads the receive ready flags from the MAC device or devices specified in the instruction and places the flags into RCV_RDY_HI register 210 a and a RCV_RDY_LO register 210 b, collectively, RCV_RDY registers 210. Each bit in these registers corresponds to a different device port on the I/O bus.
Also, and as shown in
The RCV_RDY_CNT register 216 is one of several used by the receive scheduler to determine how to issue a receive request. It also indicates whether a flow control request is issued.
There are two techniques for reading the ready bus registers: “autopush” and polling. The autopush instruction may be executed by the ready bus sequencer 160 during a receive process (rxautopush) or a transmit process (txautopush). Polling requires that a microengine thread periodically issue read references to the I/O bus interface 28.
The rxautopush operation performs several functions. It increments the receive ready count in the RCV_RDY_CNT register 216. If enabled by the RCV_RDY_CTL register 174, it automatically writes the RCV_RDY_CNT 216, the RCV_RDY_LO and RCV_RDY_HI registers 210 b, 210 a to the receive scheduler read transfer registers and signals to the receive scheduler thread 92 (via a context event signal) when the rxautopush operation is complete.
The ready bus sequencer 160 polls the MAC FIFO status flags periodically and asynchronously to other events occurring in the processor 12. Ideally, the rate at which the MAC FIFO ready flags are polled is greater than the maximum rate at which the data is arriving at the MAC ports. Thus, it is necessary for the receive scheduler thread 92 to determine whether the MAC FIFO ready flags read by the ready bus sequencer 160 are new, or whether they have been read already. The rxautopush instruction increments the receive ready count in the RCV_RDY_CNT register 216 each time the instruction executes. The RCV_RDY_CNT register 216 can be used by the receive scheduler thread 92 to determine whether the state of specific flags have to be evaluated or whether they can be ignored because receive requests have been issued and the port is currently being serviced. For example, if the FIFO threshold for a Gigabit Ethernet port is set so that the receive ready flags are asserted when 64 bytes of data are in the MAC receive FIFO 206, then the state of the flags does not change until the next 64 bytes arrive 5120 ns later. If the ready bus sequencer 160 is programmed to collect the flags four times each 5120 ns period, the next three sets of ready flags that are to be collected by the ready bus sequence 160 can be ignored.
When the receive ready count is used to monitor the freshness of the receive ready flags, there is a possibility that the receive ready flags will be ignored when they are providing new status. For a more accurate determination of ready flag freshness, the receive request count may be used. Each time a receive request is completed and the receive control information is pushed onto the RCV_CNTL register 232, the the RSM 166 increments the receive request count. The count is recorded in the RCV_RDY_CNT register the first time the ready bus sequencer executes an rxrdy instruction for each program loop. The receive scheduler thread 92 can use this count to track how many requests the receive state machine has completed. As the receive scheduler thread issues commands, it can maintain a list of the receive requests it submits and the ports associated with each such request.
Still referring to
The RCV_REQ register 230 is used to initiate a receive transfer on the Fbus and is mapped to a two-entry FIFO that is written by the microengines. The I/O bus interface provides signals (not shown) to the receive scheduler thread indicating that the RCV_REQ FIFO 236 has room available for another receive request and that the last issued receive request has been stored in the RCV_REQ register 230.
The RSM 166 reads the RCV_REQ register entry 231 to determine how it should receive data from the Fbus 132, that is, how the signaling should be performed on the Fbus, where the data should be placed in the RFIFO and which microengine thread should be signaled once the data is received. The RSM 166 looks for a valid receive request in the RCV_REQ FIFO 230. It selects the MAC device identified in the RM field and selects the specified port within the MAC by asserting the appropriate control signals. It then begins receiving data from the MAC device on the Fbus data lines. The receive state machine always attempts to read either eight or nine quadwords of data from the MAC device on the Fbus as specified in the receive request. If the MAC device asserts the EOP signal, the RSM 166 terminates the receive early (before eight or nine accesses are made). The RSM 166 calculates the total bytes received for each receive request and reports the value in the REC_CNTL register 232. If EOP is received, the RSM 166 determines the number of valid bytes in the last received data cycle.
The RCV_CNTL register 232 is mapped to a four-entry FIFO (referred to herein as RCV_CNTL_FIFO 232) that is written by the receive state machine and read by the microengine thread. The I/O bus interface 28 signals the assigned thread when a valid entry reaches the top of the RCV_CNTL FIFO. When a microengine thread reads the RCV_CNTL register, the data is popped off the FIFO. If the SIGRS field 230 i is set in the RCV_REQ register 230, the receive scheduler thread 92 specified in the RCV_CNTL register 232 is signaled in addition to the thread specified in TID field 230 k. In this case, the data in the RCV_CNTL register 232 is read twice before the receive request data is retired from the RCV_CTL FIFO 232 and the next thread is signaled. The receive state machine writes to the RCV_CTL register 232 as long as the FIFO is not full. If the RCV_CTL FIFO 232 is full, the receive state machine stalls and stops accepting any more receive requests.
The THREAD_DONE CSRs 240 support a two-bit message for each microengine thread. The assigned receive thread may write a two-bit message to this register to indicate that it has completed its task. Each time a message is written to the THREAD_DONE register, the current message is logically ORed with the new message. The bit values in the THREAD_DONE registers are cleared by writing a “1” so the scheduler may clear the messages by writing the data read back to the THREAD_DONE register. The definition of the 2-bit status field is determined in software. An example of four message types is illustrated in TABLE 1 below.
The assigned receive processing threads write their status to the THREAD_DONE register whenever the status changes. For example, a thread may immediately write 00 to the THREAD_DONE register after the receive state machine signals the assigned thread. When the receive scheduler thread reads the THREAD_DONE register, it can look at the returned value to determine the status of each thread and then update its thread/port assignment list.
The microengine supports a fast_wr instruction that improves performance when writing to a subset of CSR registers. The fast_wr instruction does not use the push or pull engines. Rather, it uses logic that services the instruction as soon as the write request is issued to the FBI CSR. The instruction thus eliminates the need for the pull engine to read data from a microengine transfer register when it processes the command. The meaning of the 10-bit immediate data for some of the CSRs is shown below.
It will be appreciated that the receive process as described herein assumes that no packet exemptions occurred, that is, that the threads are able to handle the packet processing without assistance from the core processor. Further, the receive process as described also assumes the availability of FIFO space. It will be appreciated that the various state machines must determine if there is room available in a FIFO, e.g., the RFIFO, prior to writing new entries to that FIFO. If a particular FIFO is full, the state machine will wait until the appropriate number of entries has been retired from that FIFO.
Sequenced Receive of Fast Network Port Streams of Data
A hardware sequencer wakes each thread when packet data is available for it. The thread then reads a receive control register that has information such as bus transaction status (cancel, error or fail), port ID, SOP, EOP, and byte count. Thread 0 (301) receives the first 64 byte section, which is the SOP MPKT. It reads the header portion of the SOP MPKT, verifies the header fields and uses certain of the fields to lookup the output port. The thread processing a SOP MPKT (the “SOP thread”) takes a relatively long time to process the header and perform the lookup compared the time it takes to process other types of MPKTs. When the SOP thread is done, it saves the value of the output port to a port mailbox. In one embodiment, the port mailbox is maintained in fast memory and is associated with the receive port ID.
While thread 0 is processing its data, packet data is arriving at a very high rate, faster than thread 0 can handle. The bus interface wakes thread 1 (302), which gets its receive control information. Thread 1 determines that this is a mid-packet MPKT (neither a SOP nor EOP MPKT). It need only move the packet data to memory, and then become available again. In this example, thread 0 is still processing its data.
The bus interface then wakes thread 2 (303), which gets receive control for its MPKT. Thread 2 also determines that this is a mid-packet MPKT, and moves its packet data to memory and becomes available again. Next the bus interface wakes thread 3, which is an EOP MPKT. For Ethernet packets, there can be up to 24 bus transactions from the SOP MPKT to the EOP MPKT. Thread 3 waits for the output port to be saved to the port mailbox by checking whether there is valid output port information. In one embodiment, the valid output port information is checked using a test and set, test and clear, or other asynchronous message mechanism. Thread 3 then enqueues the packet data, i.e., places the packet data in a queue to be sent to the output port.
Processing of the next data packet now begins. Thread 4 through thread 7 (305-308) behave in a fashion similar to that described with respect to threads 0 through thread 3 (301-304).
The state information is stored in a state mailbox. In one embodiment, the state mailbox is maintained in fast memory and includes 1) an address pointer, for calculating the address in DRAM where the entire packet will be stored; 2) a freelist ID, that identifies a list from which the pointer to the current packet data was obtained; 3) an element count, that is a count of the MPKT sections processed so far; 4) a packet sequence number, that is used to insure that packets are enqueued (queued up to be transmitted out of the processor) in order; and 5) a discard flag, that specifies that during processing, a thread has decided to discard the packet.
The bus interface can place a cancel message in the receive control register. If this occurs, the thread merely passes state to the next thread (proceeds from box 403 to box 405). The bus interface can also set an error bit in the receive control, in which case the thread will set the discard flag and save state to the next thread. (proceeds from box 403 to boxes 404 and 405).
In one embodiment, the processing of a packet is not complete until an EOP or FAIL is received. (FAIL is a serious error, and is treated like an EOP). If there was neither a cancel message nor an error bit set in the receive control, operation goes from box 403 to box 406. If the thread is a SOP thread, then operation proceeds to box 410, at which the thread assembles state information.
In one embodiment, the state information includes an address pointer, a freelist ID, element count, packet sequence number and discard flag, as previously described. The thread determines the address pointer, i.e., the location for the packet data in memory, and determines what freelist it got the address pointer from. There can be more than one freelist. For example, one freelist may be used to allocate memory in one DRAM bank and another freelist may be used to allocate memory in another DRAM bank. The SOP thread sets the element count to 1. Every thread thereafter will increment the element count by 1 until the EOP MPKT is received. The SOP thread also saves the packet sequence number. This number is placed in the receive control register by the bus transfer interface while the SOP MPKT is being processed. It will be used by the EOP thread to guarantee successive packets are enqueued in order. The SOP thread saves the state to the state mailbox (box 411), stores the section of packet data to DRAM (box 412), verifies the header and looks up the output port (box 413), and then saves the output port information to the port mailbox (box 414).
At box 406, if the thread is processing a mid-packet MPKT (not processing a SOP MPKT or EOP MPKT), the thread increments the element count (box 420), saves state (box 421), and stores the data section to the next section of the DRAM memory buffer (box 422).
At box 406, if the thread is processing an EOP MPKT (an “EOP thread”), then the thread increments the element count (box 430), saves a null state for the next thread (box 431) (which should be an SOP MPKT for a frame-based application, as described below), stores data to the next section of the DRAM memory buffer (box 432), and restores the output port information from the port mailbox. (In an alternate embodiment of a frame-based application, the EOP thread does not save state information to, and the subsequent SOP thread does not retrieve state information from a state mailbox.)
Note that all threads restore, update and save state as soon as possible after reading receive control. This allows overlapped threads to continually receive new packet sections at the line rate of the network. After saving state, they continue to process by moving the data to DRAM as soon as possible. This allows the bus interface buffer to be freed and used for more data streaming from the network. The thread can read the data in parallel with it being stored to memory. It then can parse, verify, modify and otherwise process the data, without impacting the receive rate. In one embodiment, a valid bit is set in a mailbox by the save operation, and is cleared by the restore operation. If, on restore, the valid bit is not set, then the restore is retried until the valid bit is set.
The embodiment for some applications is to have one mailbox per thread, i.e., each thread writes to the next thread's mailbox. Frame-based applications receive a contiguous set of packet sections starting with a SOP MPKT and ending with an EOP MPKT. A separate mailbox per thread is one way to handle frame-based applications.
Cell-based applications (e.g., ATM) receive sections of one packet interleaved with sections of another packet. A typical embodiment for these applications is to have the mailbox at a Virtual Circuit (VC) block of memory. The restore and save are to this mailbox. However, there can be many Virtual Circuits. On every packet section a Virtual Circuit lookup is performed. If on successive packet sections, the Virtual Circuit is the same, they are restored and saved to the same VC mailbox.
In one embodiment, application software for programming the receive threads to operate in the network processor as described can be provided as instructions stored on floppy disk, CD-ROM, or other storage media. Alternatively, the application software can be downloaded via the Internet. The application software is then installed to a storage medium on the host system, such as a hard disk, random access memory, or non-volatile memory.
Thus, a method of processing network data in a network processor includes using multiple receive threads to process different portions of a data packet. This scheme allows the network processor to receive data at the full line rate. However, the specific arrangements and methods described herein are merely illustrative of the principles of this invention. For example, the first mailbox could alternatively be implemented by successively storing to different physical locations instead of a static buffer area. Numerous modifications in form and detail may be made without departing from the scope of the described invention. Although this invention has been shown in relation to a particular embodiment, it should not be considered so limited. Rather, the described invention is limited only by the scope of the appended claims.